what about other datasets.
alip67 opened this issue · 5 comments
Hi, I checked the result of the paper and It seems there are more datasets that you used to compare. could you please put the code for that part as well?
Same question. How can I reproduce the results on Cora, Citeseer and Pubmed?
@alip67 what datasets are you interested in?
This repo was primarily meant for OGB submissions - we'll work on uploading the code for the rest.
However, I would advise against evaluating on Cora/Citeseer/Pubmed :P I think that we've already pushed up against the limits of those tiny citation datasets, and that evaluating on other benchmarks is more meaningful.
@Chillee First, congratulations for the acceptance of your paper. However, I try to reimplement following the idea of the C&S paper, and it does not work on the three classic small citation networks. I want to figure out what's wrong with my implement. Can u help me?
@hhr114 Thanks! The code is not very different/interesting between the different datasets. One note is that we're using "60/20/20" split setting for Cora/Pubmed/Citeseer - I don't think our method is as effective on the "public" split (i.e: 20 nodes per class), although we didn't try very hard to tune it.